Slap.KV.Telemetry (slap_kv v0.6.1)

Copy Markdown View Source

Telemetry events, all under [:slap, :kv, ...]:

  • [:write, :acknowledged] - a write was acknowledged (durable). Measurements: :duration (native time units, from when the partition writer received the request to the reply). Metadata: :shard, :op (:put, :delete or :batch).
  • [:partition_writer, :write_failed] - a write failed; the partition writer stopped and failed the requests in flight. Metadata: :shard.
  • [:request, :rejected] - a request was rejected with :overloaded (see the README's "Limits on pending work"). Measurements: :bytes (encoded keys and put values), :requests (1). Metadata: :partition, :shard, :writer (the partition writer's index, from 0 to partition_writers - 1), :op (:put, :delete, :batch, :get), :reason (:overloaded), :limit (:writer_bytes, :shard_bytes, :writer_requests, checked in that order). :request_too_large and :batch_too_large do not emit this event.
  • [:shard, :load] - the shard's pending work changed. Measurements: :bytes and :requests counted against the shard's limits, including conflicts and checks awaiting confirmation and the writes of crashed partition writers that are not yet durable. Metadata: :shard. They do not measure total memory or queued messages.
  • [:http, :request] - an HTTP response was sent. Measurements: :duration. Metadata: :method, :status.

Batch acknowledgements emit once per mutation batch. Check-only success, conflicts and linearizable gets do not emit mutation acknowledgements. The acknowledgement event is a point event without a matching :start.